| Ville Satopää†, Jeannie Albrecht†, David Irwin‡, and Barath Raghavan§
The paper "Finding a ‘Kneedle’ in a Haystack: Detecting Knee Points in System Behavior" by Ville Satopää, Jeannie Albrecht, David Irwin, and Barath Raghavan introduces Kneedle, a general approach to detecting knee points in system behavior. Knee points are critical for balancing performance and cost in computer systems, but their identification is often ad hoc and system-specific. Kneedle formally defines a knee using curvature, a mathematical concept, and evaluates its accuracy against existing algorithms on synthetic and real data sets. The authors compare Kneedle with Angle-based, Menger Curvature, and EWMA algorithms, demonstrating Kneedle's superior performance in both offline and online settings. Kneedle is applied to two real-world applications: speculative execution in MapReduce-like systems and congestion control in network protocols, showing its effectiveness in improving system performance. The paper concludes by highlighting Kneedle's utility in addressing a common problem in systems research and engineering.The paper "Finding a ‘Kneedle’ in a Haystack: Detecting Knee Points in System Behavior" by Ville Satopää, Jeannie Albrecht, David Irwin, and Barath Raghavan introduces Kneedle, a general approach to detecting knee points in system behavior. Knee points are critical for balancing performance and cost in computer systems, but their identification is often ad hoc and system-specific. Kneedle formally defines a knee using curvature, a mathematical concept, and evaluates its accuracy against existing algorithms on synthetic and real data sets. The authors compare Kneedle with Angle-based, Menger Curvature, and EWMA algorithms, demonstrating Kneedle's superior performance in both offline and online settings. Kneedle is applied to two real-world applications: speculative execution in MapReduce-like systems and congestion control in network protocols, showing its effectiveness in improving system performance. The paper concludes by highlighting Kneedle's utility in addressing a common problem in systems research and engineering.